In industrial condition monitoring, obtaining large and diverse datasets for deep learning applications, such as bearing fault diagnosis, is often challenging due to the high costs and practical difficulties. To address this issue, our study presents a novel approach that utilizes simulation-driven partial domain adaptation, based on a physics-based bearing phenomenological model. This model is instrumental in producing simulated acoustic signals, which form the basis of our domain adaptation strategy, serving as the source domain. The core of our method involves the application of contrastive learning, which effectively aligns these simulated healthy signals with real-world healthy data. This approach emphasizes partial domain adaptation and is particularly relevant in environments where faulty data is rare yet critical. By focusing on healthy data, our model adapts to the limitations of data availability. Through our in-house experimental setup, we demonstrate that leveraging simulated data in conjunction with contrastive learning for partial domain adaptation can yield promising results in fault classification, even when solely relying on healthy data. This approach offers a balanced and effective strategy for acoustic bearing fault diagnosis, presenting a significant advancement in the practical application of deep learning for industrial maintenance and condition monitoring.